论文部分内容阅读
针对不同光照以及不同天气条件造成葡萄叶片图像不能准确分割的问题,提出了一种改进的图割分割算法。采用G/R以及a*颜色特征自动选择叶片目标和背景种子点,利用混合高斯模型对叶片和背景的概率密度分布进行估计;在马尔科夫随机场的基础上,建立像素特征的能量函数;通过求解能量函数最小化对叶片实现了自动分割。对多种不同分割特征的分割效果进行对比试验,结果表明:对于不同时间、不同天气的叶片图像,单一G/R和a*具有较好的效果,分割精度分别达到86.74%和92.38%,若用它们组合为双特征,分割效果会进一步提高,分割精度可达95.03%。
Aiming at the problem that grape leaf images can not be accurately segmented due to different light conditions and different weather conditions, an improved segmentation algorithm is proposed. The leaf target and background seed points are automatically selected by using G / R and a * color features, and the probability density distributions of leaf and background are estimated by using mixed Gaussian model. The energy function of pixel features is established based on Markov random field. Automated segmentation of the blade is achieved by solving the energy function minimization. Comparing the segmentation results of different segmentation features, the results show that the single G / R and a * have good effects on leaf images at different times and different weather with the segmentation accuracy of 86.74% and 92.38%, respectively. If With their combination of double features, segmentation results will be further improved, the segmentation accuracy of up to 95.03%.